The first test: read this squiggle

The original CAPTCHA exploited one specific gap between people and machines around the year 2000: reading messy text. Your visual system is extraordinary at it — you can read letters that are warped, overlapping, crossed out and half-melted without conscious effort. The optical character recognition (OCR) software of the era? Hopeless the moment the text stopped being clean and straight.

So the trick was: take a word, distort it just enough to break OCR while leaving it readable to a human.

Scroll to crank up the distortion and watch the gap open.

Why distortion worked (then)

Old OCR assumed tidy input: neat rows of separated, upright characters. It leaned on brittle rules — find the gaps between letters, match each shape to a template. Every bit of distortion broke an assumption:

• Warping and rotation ruined the template matching.
• Overlapping letters destroyed the "find the gaps" segmentation step.
• Crossing lines and noise added fake strokes that confused everything.

Humans sail through all of this because we recognize words holistically and use context, not rigid per-letter rules. That mismatch was the CAPTCHA's entire security.

The annoying ceiling

There's a built-in tension you felt in the visual: to keep ahead of improving OCR, you have to distort more — but past a point, humans start failing too. Squint at some late-2000s CAPTCHAs and you'll remember the rage. The usable gap was shrinking from both sides, and that pressure pushed toward a completely different idea.

A wasted opportunity?

Step back and something nags. Hundreds of millions of people were squinting at words and typing them in, every day — an ocean of human effort, used once to check a box and then thrown away. What if that effort could be pointed at a real problem at the same time? That question turned the CAPTCHA from a security nuisance into one of the great data-collection engines ever built. Next: you are the labeler.